Post by Crunch Lab

A model that won one of our Crunches is now helping run cancer research in labs around the world. The open dataset it produced is downloaded 14,000 times a month on Hugging Face. Kalin Nonchev, a PhD student at ETH Zurich and one of our Crunchers, built the DeepSpot model and won the Autoimmune Disease ML Challenge we ran with the Eric and Wendy Schmidt Center, ahead of nearly a thousand researchers from 62 countries. Measuring a tumor's molecular activity directly takes special lab equipment and runs into thousands of dollars per sample. DeepSpot predicts that same activity from the cheap, standard slide a lab already has. Kalin's team ran it across 3,780 TCGA cancer slides and released the output as an open dataset: 56 million predicted gene-expression profiles spanning melanoma, kidney, and lung cancer, free for research use. The work is already cited in 2026 research across npj Artificial Intelligence, the Journal of Clinical Oncology, and Briefings in Bioinformatics. This is what Crunch is for, the place where the boldest minds bring their work to the hardest, highest-stakes problems. Credit to Kalin and his team at ETH Zurich. They brought DeepSpot to our challenge, won, and built something labs around the world now use.